Sanjay Dudani
Home  /  Analysis

Analysis

Evidence-first reads on enterprise AI.

Case files on named institutions, field guides on the work of getting AI into production, and analysis of where the industry's numbers do and do not hold up. New pieces are published here first; LinkedIn carries the summary.

How this corpus works
  1. Every figure traces to a primary source — a filing, a transcript, an on-record interview — and the source is linked.
  2. Every case file carries an independent critique: the questions the institution's own numbers do not answer.
  3. Figures that circulate widely but do not trace to a primary are named and excluded, not quietly dropped.

Case files

Named institutions, their disclosed numbers, and what a board should still ask.

Case file · Banking

JPMorgan Chase: what $2 billion of annual AI value is made of

Nine years of proprietary build, almost 1,000 use cases, 250,000 people on one generative-AI portal — and the parts of the $2 billion a board still cannot audit. Four widely repeated figures excluded for want of a primary.

JPMorgan Chase · United States · 9 September 2026
Case file · Banking

ING Bank: why 90% of its AI pilots reach production

The platform, the 20-step / 140-risk governance gate, the sequencing — and the four questions ING's own numbers do not answer.

ING Group · Netherlands · 30 August 2026

Field guides

How the work is actually done, move by move, with the evidence for each move.

Field guide · Pilot-to-production

How to move an AI pilot to production

Why most enterprise AI stalls after the demo, and the five-move sequence that gets a pilot into live operations.

Enterprises and mid-size companies · 29 August 2026
Field guide · Enablement & adoption

How to run an AI enablement program for a large team

Your people are already using AI. Whether that creates value or just risk depends on the enablement program around it.

Large organisations · 29 August 2026
Field guide · Go-to-market

How to build an AI-led go-to-market system

For a startup, go-to-market is the whole game — and AI makes it possible to build GTM as a system from day one instead of hiring your way to scale. And the trap most teams fall into.

Startups and scale-ups · 29 August 2026

Analysis

Where the industry's numbers hold up, where they do not, and what that means for an operator.

Analysis · Agentic AI

Shipping agents before the audits exist: why AI agents fail in production

The constraint is audit architecture, not model quality. The same model moves from 86% to 99% reliability when verification is built around it. Eleven sources, all primary.

Agentic AI · 30 August 2026

Want this read on your own programme?

A small number of advisory engagements a year: pilot-to-production, AI enablement and AI-led go-to-market. The first step is an honest diagnosis, in writing.

Request a conversation